File size: 2,288 Bytes
8b7c501 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 |
// Auto-generated file. Do not edit!
// Template: src/f32-gemm/sse-load1.c.in
// Generator: tools/xngen
//
// Copyright 2019 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
#include <assert.h>
#include <immintrin.h>
#include <xnnpack/gemm.h>
void xnn_f32_gemm_minmax_ukernel_1x8__sse_load1(
size_t mr,
size_t nc,
size_t kc,
const float* restrict a,
size_t a_stride,
const float* restrict w,
float* restrict c,
size_t cm_stride,
size_t cn_stride,
const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
assert(mr != 0);
assert(mr <= 1);
assert(nc != 0);
assert(kc != 0);
assert(kc % sizeof(float) == 0);
assert(a != NULL);
assert(w != NULL);
assert(c != NULL);
const float* a0 = a;
float* c0 = c;
do {
__m128 vacc0x0123 = _mm_load_ps(w + 0);
__m128 vacc0x4567 = _mm_load_ps(w + 4);
w += 8;
size_t k = kc;
do {
const __m128 va0 = _mm_load1_ps(a0);
a0 += 1;
const __m128 vb0123 = _mm_load_ps(w);
const __m128 vb4567 = _mm_load_ps(w + 4);
w += 8;
vacc0x0123 = _mm_add_ps(vacc0x0123, _mm_mul_ps(va0, vb0123));
vacc0x4567 = _mm_add_ps(vacc0x4567, _mm_mul_ps(va0, vb4567));
k -= sizeof(float);
} while (k != 0);
const __m128 vmax = _mm_load_ps(params->sse.max);
vacc0x0123 = _mm_min_ps(vacc0x0123, vmax);
vacc0x4567 = _mm_min_ps(vacc0x4567, vmax);
const __m128 vmin = _mm_load_ps(params->sse.min);
vacc0x0123 = _mm_max_ps(vacc0x0123, vmin);
vacc0x4567 = _mm_max_ps(vacc0x4567, vmin);
if XNN_LIKELY(nc >= 8) {
_mm_storeu_ps(c0, vacc0x0123);
_mm_storeu_ps(c0 + 4, vacc0x4567);
c0 = (float*) ((uintptr_t) c0 + cn_stride);
a0 = (const float*) ((uintptr_t) a0 - kc);
nc -= 8;
} else {
if (nc & 4) {
_mm_storeu_ps(c0, vacc0x0123);
vacc0x0123 = vacc0x4567;
c0 += 4;
}
if (nc & 2) {
_mm_storel_pi((__m64*) c0, vacc0x0123);
vacc0x0123 = _mm_movehl_ps(vacc0x0123, vacc0x0123);
c0 += 2;
}
if (nc & 1) {
_mm_store_ss(c0, vacc0x0123);
}
nc = 0;
}
} while (nc != 0);
}
|